My recommendation: Before scaling AI across plants, I recommend stabilizing the process it will accelerate. Define the work, required inputs, handoff conditions, exception ownership, and evidence of reliable output. Then test one bounded application against quality, rework, delivery, and review effort before deciding whether other sites are ready.
During my time working with Toyota, I learned a principle that still shapes how I help companies today: you have to create stability before you can increase capability.
Now, as the CEO of Incito, I visit organizations doing $250 million to $1 billion in revenue. Leaders walk me through their operations, show me their technology, and explain the AI systems they are preparing to implement. I understand their excitement. But when the underlying process is inconsistent, I have to pause the conversation.
Are we stable enough to implement this system? What happens if we accelerate the problems already here?
Before you add AI to a process, make certain that process is stable. Otherwise, you risk producing waste faster. The place to start is the work itself.
If you are a COO, plant leader, or continuous improvement leader preparing to scale AI across operations, I would start with one value stream. Before you replicate a pilot across plants, establish which methods must be consistent and which conditions genuinely differ by site. A successful demonstration in one area does not establish readiness across the enterprise.
How do you assess a manufacturing process before adding AI?
I start by following one real order or production job across the functions that deliver it. Compare the documented method with actual practice. A process readiness review should reveal unclear inputs, inconsistent decisions, and unresolved handoffs before software spreads those conditions across more work.
- Is there a clear way to do the work?
- Do people follow it consistently?
- When something goes wrong, can they identify the problem and correct it?
Those questions sound basic. They are basic. But the answers tell you more about readiness than a demonstration of the technology ever will.
Follow an actual order, quote, service request, or production job through the process. Look at the information people receive, the decisions they make, and the work they hand to the next person. Compare the documented method with what happens on the floor or in the office.
If the process depends on someone knowing whom to call, which spreadsheet to trust, or how to work around an unclear instruction, you need to understand that before you automate it.
What needs to be standardized before an AI rollout?
Before an AI rollout, I want a clear baseline method, dependable inputs, agreed completion conditions, and an owner for exceptions. Production standardized work also addresses timing, work sequence, and the inventory required for flow. Establish the baseline so the team can recognize when performance changes.
- Agree on the information required to begin.
- Define the sequence and the conditions for a complete handoff.
- Make ownership of problems clear.
Standardized work gives us a baseline for improvement. In production, that includes timing, sequence, and the inventory needed to support flow. In administrative work, we still need a clear method, reliable inputs, and an agreed definition of completion.
I am not suggesting that every process must be perfect before you use AI. I am asking whether you know what the process produces, where it varies, and how the team responds when it encounters something outside the normal conditions.
Consider a growing manufacturer introducing AI into order release across several plants. This is an illustrative application: sales, planning, production, and shipping need the same definition of a complete order. I would compare exceptions by site and check whether the change supports reliable delivery at the higher volume before expanding it.
When is an AI pilot ready to scale across plants?
I would expand an AI pilot only after the team can show that the process works reliably and receiving sites can support it. Review quality, rework, delivery, human checking effort, and local differences. Agree on release conditions, an accountable owner, and a way to pause if performance deteriorates.
- Begin with one repeatable task and a clear output.
- Compare quality and rework alongside elapsed time.
- Keep a review point and a practical way to pause.
Once the method is clear, test whether AI improves it. Look beyond the time saved at one step. Does someone else now spend more time checking the output? Are errors reaching the next department sooner? Is the customer receiving a more reliable result?
A value stream review helps you see those connections. The value of the technology has to show up in the work and in what the customer receives.
Before you invest in making a process faster, understand what you will be making faster. Create stability, then build capability.
Frequently asked questions
Does every process need to be perfect before using AI?
No. I want the team to understand its current method, its known exceptions, and how it will recognize a problem. That gives you a sound basis for a limited, controlled trial.
What should I measure besides speed?
Look at quality, rework, exception handling, and the effect on the customer. Include the human effort needed to check or correct the output.
How should we decide whether an AI pilot is ready for other plants?
I would check the method, input quality, exception ownership, and delivery results at the pilot site, then compare the receiving sites. Set explicit release conditions and a local owner; do not assume that matching software means matching operating conditions.
Sources
Assess your operations before scaling AI
If AI investment is part of your capacity or multi-site growth plan, let’s examine the process stability, handoffs, and leadership controls needed before rollout.
Discuss operational readiness for AI